Introduction

Currently, esophagectomy within a multimodal treatment plan is the preferred management of patients with resectable esophageal cancer [16]. The introduction of minimally invasive techniques for esophagectomy has revolutionized surgical treatment leading to lower perioperative morbidity and better quality of life [18, 21, 26]. Hybrid laparoscopic/thoracoscopic minimally invasive esophagectomy (MIE) and robotic-assisted minimally invasive esophagectomy (RAMIE) have both led to a significant reduction in pulmonary infections and postoperative pain in randomized clinical trials [2, 19, 34] while maintaining oncologic radicality [24]. Some key advantages of the robotic-assisted technique, especially during transthoracic resection and reconstruction, are an increased range of motion of the instruments within the rigid thoracic cage, the optional use of three arms, and an improved surgical view with standard 3DHD visualization [13]. Although several techniques of reconstruction after MIE have been reported, the majority of European centers favor minimally invasive intrathoracic esophagogastrostomy [14, 35]. The use of a circular stapler appears to be advantageous with regard to the AL rate, although this question has not yet been conclusively clarified [5]. Experienced centers have published the first larger single-center reports of RAMIE with excellent oncological results and low mortality rates of 1–3% [25, 33].

The German da Vinci ** of 15 consecutive patients was performed for each center. Moreover, we accomplished a case grou** (n=15) for each surgeon with more than 30 cases and studied the median reduction in the operative time of subsequent cases compared with the initial 15 cases. For further investigation of the learning curve, a cumulative sum (CUSUM) analysis of the total operative time was performed. This technique is a graphical method to transform raw data into a running total of differences from the group average. Therefore, a chronological arrangement of all cases from the first to the last by the center (or by the leading surgeon, respectively) was performed. Then, CUSUM values were calculated according to the following formula: CUSUM = Σ (x i−μ), where x i is the total operative time of the individual case and μ is the mean operative time of the corresponding center or leading surgeon [30, 36]. Finally, the CUSUM values were plotted on the vertical axis according to their case number on the horizontal axis. Learning curves could be determined by visual interpretation of the chart. The end of the learning curve was predefined as inflection of the curve to a plateau or decrease.

Continuous variables are presented as medians with interquartile ranges (IQRs). The evaluation for nonparametric variables was performed with the Mann-Whitney U test. Univariate analysis was computed using cross tabulation and the chi-square test or Fisher’s exact test.

Results

Patient characteristics and histopathological results

In total, 220 patients were included in the analysis (center 1: 72; center 2: 41; center 3: 83; center 4: 10; center 5: 14). The median age of the patients was 64 years (IQR 58–72), and 85.5% (n=188) were male. Two-thirds of the patients had significant comorbidities (ASA ≥III), and the median BMI was 26.2 kg/m2 (IQR 23.6–29.4). No further information of race or ethnicity of the patients was collected routinely in the **s (p=0.024). The median operative time subsequently dropped from 439 min (cases 16–30; n=45) to 402 min (cases 31–45; n=41) (p=0.126). Although the median operative time was further reduced to 349 min after >60 cases (n=35), the difference was not statistically significant when compared with cases 46–60 (p=0.089; n=30) (Fig. 2a). Similar results were observed if the median operative time of the thoracic part only was analyzed: the median operative time of cases 1–15 was longer than that of the consecutive case grou**s (p=0.023). However, a significant reduction in the operative time for the abdominal part was not seen until a caseload >60 (p≤0.001) (Suppl. Fig. 1). The three most experienced surgeons of the participating centers could significantly improve their individual median operative time by approximately −4.8% after cases 16–30 (n=45) and approximately −11.6% after >30 cases (n=50) (p≤0.021). Likewise, there was a trend toward a further operative time reduction from cases 16–30 to cases >30, but without statistical significance (p=0.057) (Fig. 2b).

Fig. 2
figure 2

Operative time of the RAMIE procedure. A Operative time including abdominal and thoracic parts of all procedures in all 5 centers (n=220). The median operative time (min) is shown stratified by chronological grou** of 15 cases (*p≤0.024). B Operative time for the three surgeons with >30 RAMIE procedures. The graph displays the median difference in the operative time from the first 15 cases compared with cases 16–30 and cases >30 for the three surgeons with the highest case load (*p≤0.021)

Full size image

The pooled CUSUM graph for all centers showed a peak (inflection point) with a slow decrease after 22 cases, indicating the end of the learning curve for the total operative time (Fig. 3a). The CUSUM graphs for the three centers with more than 22 RAMIE procedures revealed different end points of the learning curve: the inflection point in center 1 was at 22 cases, center 2 reached a plateau after 13 cases, and center 3 reached a plateau after just 10 cases (Fig. 3b). The CUSUM analysis for the leading surgeons of the three most experienced centers showed the same end points of the learning curve of surgeons B and C as for their related centers 2 and 3 (Fig. 3c). This finding is not surprising because the leading surgeon in these two centers performed (nearly) all procedures (78.3% and 100%, respectively). In center 1, three surgeons routinely performed RAMIE, which explains the longer learning curve for this center. The point of inflection for leading surgeon A of center 1 was at the 9th case (Fig. 3c).

Fig. 3
figure 3

CUSUM analysis of the operative time. A CUSUM analysis including all five centers. The inflection point after the 22nd procedure marks the end of the learning curve. B CUSUM analysis including the three centers with >22 RAMIE cases. C The CUSUM analysis for the leading surgeons of the three most experienced centers

Full size image

The impact of the learning curve on intraoperative findings, postoperative course, and mortality

Based on the CUSUM analysis, perioperative outcome parameters were compared in relation to the pre- and post-learning curve cohort (≤22 and >22 cases). The latter cohort was operated on with less blood loss (p<0.001), a shorter operative time (p<0.001), and a lower rate of postoperative pneumonia (p=0.046). Additionally, there was a trend toward a lower conversion (11.1 to 4.6%; p=0.061) and 90-day readmission rate (12.2to 5.4%, p=0.059) after >22 cases. Other outcome parameters, including major complications CDC ≥3b, AL rate, textbook outcome, and intensive care parameters, were not significantly different between the two groups (Table 4).

Table 4 Intra- and postoperative findings dependent on case number (n=220)
Full size table

All operated cases (n=220) were included in a univariate analysis to identify predictive factors for the development of AL (Suppl. Tab. S1). However, none of the tested variables significantly correlated with the occurrence of AL.

Discussion

This is the first report of a prospective multicenter registry trial evaluating the short-term outcome of RAMIE with an intrathoracic circular stapled anastomosis. Because of the potentially beneficial effects of the RAMIE approach on short-term patient outcome, the participating university centers agreed on a prospective multicentric registry study to evaluate the safety and potential benefits of RAMIE during the implementation phase and beyond with a standardized technique. The aim of this registry study was to generate data to assess the da Vinci ** surgical system for esophagectomy regarding clinical outcome. The multicenter RAMIE program included a uniform technique with an intrathoracic (Ivor Lewis) circular stapled esophagogastrostomy using a minithoracotomy. According to the present knowledge, circular stapled anastomosis is the most frequently performed anastomosis technique during RAMIE [14].

Overall, approximately 80% of the operations were minimally invasive using the da Vinci ** robotic system (fully robotic), and the thoracic part, including the anastomosis, was robotic-assisted in 94% (207) of the cases; whereas in thoracoscopic approaches (MIE), the conversion rate was 14% [2, 28]. This result demonstrates that the RAMIE technique is feasible in most cases. The rate of a fully robotic-assisted approach was higher than that in a recent international registry report, where only 54% of the cases were not hybrid procedures [14]. Other high-volume centers for RAMIE combine an abdominal open or laparoscopic part with the robotic-assisted thoracic part [25]. A direct comparison of an open abdominal operation phase and a total RAMIE revealed no significant differences regarding oncological radicality and recurrence-free survival, suggesting that robotic-assisted abdominal lymphadenectomy is adequate [23]. The oncological quality in the present study, as indicated by the R0 resection rate (93%) and the median number of resected lymph nodes (n=25), is comparable with recent single-center series [14, 25, 34].

The current results from leading esophagus surgery centers support the use of a circular stapler esophagogastrostomy, especially for minimally invasive intrathoracic anastomosis [5, 22]. The stapler diameter should be selected according to the individual anatomical situation of the patient but with preference for the largest possible diameter; however, a significant difference regarding anastomotic leakage and stricture was not identified between 25- and 28-mm diameter sizes [29]. In our analysis, in 84% of all cases, a stapler size equal to or greater than 28 mm was used for esophagogastrostomy, which could contribute to the markedly low leakage rate.

According to the available randomized data, the strength of MIE/RAMIE is the lower postoperative morbidity, especially a reduced rate of pulmonary complications. A recent propensity score-matched comparison and meta-analysis concluded that RAMIE has significantly lower rates of pneumonia or pulmonary complications than laparoscopic MIE and should potentially be considered the standard technique for esophagectomy [31, 37]. In the present trial, the rates of postoperative pneumonia and anastomotic leakage were 19.5% and 13.2%, respectively, compared with 23% and 20%, respectively, in the international registry (out of the 331 fully robotic Ivor Lewis cases) [14].

Interestingly, the present analysis showed that key characteristics and complications such as operative time, blood loss, the rate of pneumonia, and anastomotic leakage can be further improved after a learning experience of 22 cases, which was the initial CUSUM-based learning curve plateau for all five centers. The CUSUM analysis was designed for detecting minor changes in datasets to visualize trends describing the learning curve [8]. Interestingly, in another German single-center analysis, also a case load of 22 was necessary to overcome the learning curve for RAMIE procedure [1]. A comparable number of cases for completion of the learning curve (20–24 cases) have been reported by other centers, especially for experienced robotic-assisted surgeons [10, 15, 32]. In contrast, MIE was usually coupled with longer learning processes with flat learning curves; 54–119 cases were reported to be required to reach a stable plateau [3]. Robotic-assisted surgery instead displays distinct steeper learning curves, likely due to special da Vinci surgical system training programs and the existing competence of most participating surgeons in MIE surgery [27]. The present study further confirms that single experienced surgeons can reach the plateau for RAMIE within a proctored program even earlier.

Prior experience in robotic-assisted surgery seems to be of high importance. Increased overall and pulmonary complications and reoperation rates were observed after the TIME trial setting, with excellent short-term outcomes after MIE implementations in nationwide practice. The authors concluded that this may reflect the completion of the MIE procedure by nonexpert surgeons in a nonstandardized fashion outside of high-volume centers [20]. Therefore, recommendations toward RAMIE should be given after considering the center volume and experience of the leading surgeons.

The advantage of the study design is the multicenter setting with a uniform technique and the high quality of the data that was prospectively recorded and closely monitored. Alternatively, the data are limited by a heterogeneous set of lead surgeons and assistants in different centers, and minor modifications of the standard operative techniques were observed (e.g., insertion of jejunal feeding tubes or differences in the number of chest drains or the use of oral antibiotics on the day before the operation).

Conclusions

In conclusion, the present high-quality multicenter registry data confirm that RAMIE is a safe procedure and can be reproduced with acceptable leak rates and promising short-term results in a multicenter setting. The learning curve is comparably low at approximately 22 cases for experienced surgeons and in a setting with interinstitutional proctoring.